AI customer service for ecommerce: what it can and cannot do

What AI customer service means for a small store

AI customer service is software that reads a shopper's message and writes the answer, either on its own or as a draft for you to send. On an online store it usually sits in a chat window on the site, in your email inbox, or in the direct messages of your social accounts.

Tools work in one of two ways. An assistant suggests replies and you press send. An AI agent answers the shopper directly and hands the conversation to you when it is unsure or when the shopper asks for a person. Starting with suggested replies means nothing reaches a customer before the owner has looked at it.

Either way, the tool is only as good as what it can read. A chatbot connected to your order system can tell a shopper where a parcel is. A chatbot that only knows your product pages can describe a candle's burn time but cannot see that the order shipped yesterday.

What an ecommerce chatbot does well

Small stores get the same few questions again and again. Give an ecommerce chatbot the questions with one right answer that is already written down, and keep anything that needs a decision.

• Order status. With access to orders, the bot can look up an order by email or order number and share the tracking link.
• Shipping questions. Delivery times, the countries you ship to and the cost of shipping, taken from your shipping settings.
• Return and exchange steps. How to start a return, the time window and the condition items must be in, quoted from your written policy.
• Product details. Sizes, materials, care instructions and ingredients, taken from your product pages.
• After-hours cover. Shoppers who write at night get an answer or a clear note that you will reply in the morning.
• Drafting. For anything unusual, the assistant writes a first draft that you edit, which is faster than starting from a blank reply.

What AI customer service cannot do

AI models write fluent text, and fluent text can be wrong. A chatbot can state a refund rule you never set, promise a delivery date your carrier will not meet, or describe a product feature that does not exist.

• It cannot see the parcel. Damaged, missing or wrong items need photos, a carrier claim and a decision about a refund or a reshipment.
• It should not make judgement calls. Whether to refund outside your return window, or give a discount after a bad experience, is a decision about your money and your customer.
• It does not know what you have not written down. If your returns policy says nothing about sale items, the bot may guess.
• It cannot calm every angry customer. Some shoppers want a person, and a bot that keeps answering makes it worse.
• It is not a safe source of medical, safety or legal claims. Questions about allergies, skin reactions or whether a product is safe for children need a careful human answer that matches your labels.

AI chatbot for ecommerce: how four tools charge

The four tools below show two ways to charge for AI customer service. Shopify Inbox is a free app. The other three charge for each conversation the AI handles on its own, by their own definition of a resolution or an outcome, usually on top of a helpdesk plan. That keeps the cost low in a quiet month and raises it in a busy one.

The table shows how each tool charges and does not rank them. Each tool counts a resolved conversation in its own way, so read that definition before you compare.

The AI fee is not the whole bill. On Intercom's own plans, you pay per teammate based on your plan, plus usage such as Fin outcomes. Zendesk AI agents are included in every Suite and Support plan, with pricing based on the successful outcomes they deliver.

Before you pick a per-resolution tool, count the support messages you got last month and how many were repeat questions. Example: a store gets 200 messages a month and 100 of them are order-status questions. If the AI resolved every one of those, the store would pay for 100 resolutions that month.

On a Shopify store, you can test Inbox before paying for anything: it is free and learns from your catalog, policies, and store content while you control its tone, style, and rules.

Rules that apply when a bot answers your customers

A chatbot speaks for your store, so treat what it tells a shopper as something you said yourself. When the US Federal Trade Commission announced Operation AI Comply on September 25, 2024, its chair said there is no AI exemption from the laws on the books.

Rules vary by state, city and country, and they change. This section is general information, not legal advice. If you sell into many places or handle sensitive products, ask a lawyer who knows your market.

• Shipping promises. Under the FTC's Mail, Internet, or Telephone Order Rule, you must have a reasonable basis for stating or implying that you can ship within a certain time. The FTC's guide adds that the updated shipment information you provide on the telephone or the Internet supersedes any shipment representation you made in the advertising, and you also must have a reasonable basis for the updated shipment representation. A bot that tells shoppers when an order will ship is making those statements for you, so it should only repeat the times in your shipping settings.
• Saying it is a bot. California's Business and Professions Code section 17941 makes it unlawful for any person to use a bot to communicate or interact with another person in California online, with the intent to mislead the other person about its artificial identity for the purpose of knowingly deceiving the person about the content of the communication in order to incentivize a purchase or sale of goods or services in a commercial transaction. The same section says a person using a bot shall not be liable under this section if the person discloses that it is a bot, and the disclosure must be clear, conspicuous, and reasonably designed to inform persons with whom the bot communicates or interacts that it is a bot.
• Wrong answers are still your answers. In Moffatt v. Air Canada, 2024 BCCRT 149, a tribunal in British Columbia looked at an airline chatbot that suggested Mr. Moffatt could apply for bereavement fares retroactively, although Air Canada did not permit retroactive applications. The tribunal said the airline should know it is responsible for all the information on its website, that it makes no difference whether the information comes from a static page or a chatbot, and that the airline did not take reasonable care to ensure its chatbot was accurate. The case was decided in Canada, but it shows how a dispute over a bot's answer can go.

How to set up AI customer service on a small store

These steps work for any tool, built in or added as an app.

1. Write down your policies

   Put shipping times, the countries you ship to, the returns window, exchange rules and what happens with sale items on clear pages. The bot answers from these pages, so gaps become guesses.

2. Collect your most common questions

   Read your last month of emails and messages and list the questions that repeat. Write one short, correct answer for each and put it on your FAQ or policy pages, or wherever your tool reads its answers from.

3. Start with suggested replies

   Let the AI draft and you send for the first few weeks. You see where it is right, where it is vague and where it is wrong before a shopper does.

4. Set hand-over rules

   Send refunds, damaged items, safety questions and any request for a person straight to you. Tell shoppers when you will reply.

5. Tell shoppers they are talking to a bot

   Label the chat as automated and offer a way to reach a person. It sets fair expectations, and the rules section above covers what California's bot law says about disclosure.

6. Read the transcripts every week

   Check a sample of conversations the bot closed on its own. Fix the policy page or saved answer behind every wrong reply.

How to tell if it is working

Look at three things each week: how many conversations the AI resolved without you, how many it handed over, and how many it got wrong. A falling count of repeat questions in your own inbox is a good sign it is helping.

Watch for shoppers who ask the same question twice in one conversation, or who ask for a person after the bot has answered. Both mean the answer was unclear or wrong. If a pay-per-resolution tool counts conversations as resolved that shoppers then reopen by email, you are paying for answers that did not help.

Where Nation fits

Questions about AI customer service